作者/Author(s): Ruby Scanlon 

網站來源/Source: Lawfare 

日期/Date: February 14, 2025 

關鍵字/Keywords: 科技、DeepSeek、中國、AI 


摘要：
在中國推出 DeepSeek（與 OpenAI 的 GPT-4 對應的產品）之後，美國科技公司陷入恐慌。在美國對先進晶片的出口管制下，中國公司是如何成功的？ 
  • 中國政府高度重視並大力投資 AI 發展，確保中國能在全球 AI 競爭中佔有一席之地。 
  • DeepSeek 的母公司原是一家專注於 AI 股票交易的避險基金，但在北京政府的監管整頓後轉向 AI 研究。 
  • 浙江省政府指定 DeepSeek 為「國家高新技術企業」，並給予其更多獎勵。DeepSeek地位的提升凸顯了北京對發展本土人工智慧的重視，以及人工智慧在其科技政策中的重要性。 
  • 自 2017 年以來，中共將人工智慧視為國家權力的必要因素。因此，他們鼓勵民營企業和地方政府開發人工智慧，積極推動政府與私營部門的合作。 
  • DeepSeek 推出後，中國銀行（BoC）隨即宣布投資 AI 領域，促進中國科技自給自足。 
  • 鑒於美國實施出口管制，中國必須確保國內運算基礎設施的安全。因此，他們得使用比美國更少的硬資源來運作。 
  • 為此，各省政府補貼 AI 研究實驗室，以減輕企業的雲端運算和數據處理成本。中央政府也在中國西部建設新數據中心，向戰略性產業提供資金支持。然而，這些努力仍無法完全替代高階 AI 晶片，中國 AI 企業仍需依賴舊型 AI 晶片。 
  • 北京政府鼓勵 AI 企業之間的競爭，並透過補助計畫扶植 AI 企業發展。各省政府相互競爭，推動 AI 創新。 
  • 然而，大量調動資源進行人工智能開發可能會導致效率低下、投資失當和市場扭曲。然而北京的大力扶持仍促成了DeepSeek 的成功。 
  • DeepSeek 的突破令美國政策制定者震驚，但美方應從中國的 AI 產業發展策略中汲取經驗，特別是將企業利益結合國家戰略目標、投資於本土優秀企業，並透過國內市場競爭促進 AI 創新。 
  • 雖然美國仍擁有全球頂尖的 AI 企業，但卻缺乏高品質數據與人工智能能源供需，恐無法長期維持領導地位。中國的「國家主導」模式可能是解決美國 AI 發展困境的最佳方案。 
 
Summary: 
US technology companies panicked after China launched DeepSeek, its indigenous counterpart to OpenAI's GPT-4. How did a Chinese company succeed despite export controls on advanced chips? 
  • The Chinese government prioritized and invested in AI development, ensuring China has a globally competitive AI. 
  • DeepSeek's parent company started as a hedge fund for AI stock trading but transitioned into AI research after the Beijing government cracked down on it. 
  • The Zhejiang provincial government designated DeepSeek a "national high-tech enterprise," giving it more incentives. DeepSeek's elevated status signified Beijing's focus on developing indigenous AI and its importance in its technology policies. 
  • Since 2017, the CCP has perceived AI as essential to national power. Consequently, China has encouraged private firms and local governments to develop AI and promote government-private sector cooperation. 
  • After DeepSeek's introduction, the Bank of China pledged investment to promote China's science and technology self-sufficiency. 
  • Due to US export controls, China must secure domestic computing infrastructure. Therefore, China must operate with fewer hard resources than the US. 
  • Provincial governments subsidized AI labs to reduce the burden of cloud computing and data processing. The central government also built new data centers in China's western provinces and provided funds for strategic sectors. However, these efforts could not replace access to advanced chips, as Chinese firms still rely on older models of AI chips. 
  • Beijing also encouraged competition among AI companies and provided grants to nurture AI firms. Provincial governments also competed against one another to drive AI innovation. 
  • However, heavy mobilization of resources for AI development could cause inefficiencies, malinvestment, and market distortion. Nevertheless, the Beijing government's efforts bore fruit through the success of DeepSeek. 
  • DeepSeek's breakthrough startled US policymakers, but they should learn from China's approach to developing indigenous AI, particularly in aligning company interests with national priorities, investing in local champions, and promoting domestic competition to foster AI innovation. 
  • Although the US is still home to top AI companies, it may not sustain its leadership for long because it is short of high-quality data and faces AI energy supply and demand issues. China's state-centric approach is the best solution to the US's woes. 
 
回上層